p5pRT
commented
21 years ago From stephens@ion09.dev.ionink.com
Created by stephens@ion09.dev.ionink.com
I have a set of Perl modules that cause memory corruption and subsequent SIGSEGV. This bug seems to go back to at least Perl 5.6.1.
At first I was using a stock perl5.6.1 and perl5.8.0 on RedHat 8.0. They were Configured without -Dusemyalloc\, this was crashing in a libc malloc() internal function\, in both perl5.6.1 and perl5.8.0\, during parsing. I recompiled perl5.8.0 with -Dusemymalloc and -Doptimize='-g'\, to rule out problems with libc malloc().
Here is the stacktrace from gdb:
====================================================================
Current directory is /tmp/stephens/perl-5.8.0-debug/src/perl-5.8.0/ GNU gdb Red Hat Linux (5.2.1-4) Copyright 2002 Free Software Foundation\, Inc. GDB is free software\, covered by the GNU General Public License\, and you are welcome to change it and/or distribute copies of it under certain conditions. Type "show copying" to see the conditions. There is absolutely no warranty for GDB. Type "show warranty" for details. This GDB was configured as "i386-redhat-linux"... /tmp/stephens/perl-5.8.0-debug/src/perl-5.8.0/test4.pl: No such file or directory. (gdb) cd /home/stephens/ion/src/uml2code/test Working directory /home/stephens/ion/src/uml2code/test. (gdb) run test4.pl Starting program: /tmp/stephens/perl-5.8.0-debug/src/perl-5.8.0/perl test4.pl
Program received signal SIGSEGV\, Segmentation fault.
0x080698ba in Perl_gv_check (stash=0x8247b80) at gv.c:1153
(gdb) bt
#0 0x080698ba in Perl_gv_check (stash=0x8247b80) at gv.c:1153
#1 0x08069915 in Perl_gv_check (stash=0x823a850) at gv.c:1157
#2 0x08069915 in Perl_gv_check (stash=0x816108c) at gv.c:1157
#3 0x0805fe4d in S_parse_body (env=0x0\, xsinit=0x805caa0 \<xs_init>) at perl.c:1548
#4 0x0805efec in perl_parse (my_perl=0x815ff08\, xsinit=0x805caa0 \<xs_init>\, argc=2\, argv=0xbfffe0e4\, env=0x0)
at perl.c:1041
#5 0x0805ca52 in main (argc=2\, argv=0xbfffe0e4\, env=0xbfffe0f0) at perlmain.c:83
#6 0x420158d4 in __libc_start_main () from /lib/i686/libc.so.6
(gdb) up
#1 0x08069915 in Perl_gv_check (stash=0x823a850) at gv.c:1157
(gdb) up
#2 0x08069915 in Perl_gv_check (stash=0x816108c) at gv.c:1157
(gdb) p hv
$1 = (struct hv *) 0x823a850
(gdb) p *hv
$2 = {sv_any = 0x8236468\, sv_refcnt = 4\, sv_flags = 536870923}
(gdb) p hv->sv_any
$3 = (struct xpvhv *) 0x8236468
(gdb) p *(hv->sv_any)
$4 = {xhv_array = 0x81d9c08 ""\, xhv_fill = 3\, xhv_max = 7\, xhv_keys = 3\, xnv_nv = 0\, xmg_magic = 0x0\,
xmg_stash = 0x0\, xhv_riter = -1\, xhv_eiter = 0x0\, xhv_pmroot = 0x0\, xhv_name = 0x81fe628 "UML"}
(gdb) do
#1 0x08069915 in Perl_gv_check (stash=0x823a850) at gv.c:1157
(gdb) p *(hv->sv_any)
$5 = {xhv_array = 0x8166e08 "\003\003\003\003\003\0034\v (\022\v"\, xhv_fill = 13\, xhv_max = 31\, xhv_keys = 16\,
xnv_nv = 0\, xmg_magic = 0x0\, xmg_stash = 0x0\, xhv_riter = -1\, xhv_eiter = 0x0\, xhv_pmroot = 0x0\,
xhv_name = 0x82c5128 "UML::MetaMetaModel"}
(gdb) do
#0 0x080698ba in Perl_gv_check (stash=0x8247b80) at gv.c:1153
(gdb) p entry
$6 = (struct he *) 0x3030303
(gdb) p (char)0x30
$7 = 48 '0'
(gdb) l
1148
1149 if (!HvARRAY(stash))
1150 return;
1151 for (i = 0; i \<= (I32) HvMAX(stash); i++) {
1152 for (entry = HvARRAY(stash)[i]; entry; entry = HeNEXT(entry)) {
1153 if (HeKEY(entry)[HeKLEN(entry)-1] == ':' &&
1154 (gv = (GV*)HeVAL(entry)) && (hv = GvHV(gv)))
1155 {
1156 if (hv != PL_defstash && hv != stash)
1157 gv_check(hv); /* nested package */
(gdb) p i
$8 = 0
(gdb)
====================================================================
The first entry of the "stash" hash array in the "UML::MetaMetaModel" stash appears to be corrupted\, which send "entry" into outer space to a planet filled with ASCII '0' ("zero"\, not null) characters.
The script "test4.pl" is part of a UML meta-meta-model toolkit\, "uml2code"\, I am working on. It generates a Perl (\, XMI or Java) implementation of the UML meta-model by bootstrapping from with a meta-meta-model\, which is a subset of the meta-model. Whew!
You can download uml2code at http://kurtstephens.com/pub/uml2code/release/uml2code-20030417.tar.gz
I have not been able to get it to happen with a smaller code set. The failure maybe related to the really hairy Data::Dumper code in uml2code/test/gen/UML/MetaModel.pm.
Prerequisites for uml2code are Parse::RecDescent and XML::Writer.
Something related: If you run perl -DT test4.pl it doesn't SIGSEGV\, it fail with "Can't call method "generalization" on unblessed reference at ../lib/perl/UML/MetaMetaModel/Util.pm line 269." (after a really long time; go eat lunch). This die() is probably my fault.
Please let me know how I can help.
Thanks\, Kurt Stephens ks.perl@kurtstephens.com
====================================================================
Migrated from rt.perl.org#21982 (status was 'resolved')
Searchable as RT21982$